It has a good noise immunity to take the high-order cummulants of the signal as it has characteristic parameters. However, different modulation type may have the same characteristic parameters, which result in partial identified.so, it is necessary to bring in fractal theory. The paper takes the cmbination of the high-order cummulants and the fractal dimension as the characteristic parameters of the signal.it not only avoid the Gaussian noise impact on the modulation recognition, but also solved the partial identified problem exist in the situation that take the order cumulants as the single characteristic parameters. Simulation results show the effectiveness of the algorithm.